Installation error with cuda

[YingchangLiu]

Hi Alan, thanks for your deepwave.
Deepwave-cpu works well for me. But when I install deepwave with cuda on my windows, it returns an error. Looking for your help.
Here is the report from pip.

 "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.6\bin\nvcc" -c deepwave\scalar\scalar_gpu.cu -o build\temp.win-amd64-3.7\Release\deepwave\scalar\scalar_gpu.obj -Ideepwave\scalar -ID:\Miniconda3\envs\deepwave\lib\site-packages\torch\include -ID:\Miniconda3\envs\deepwave\lib\site-packages\torch\include\torch\csrc\api\include -ID:\Miniconda3\envs\deepwave\lib\site-packages\torch\include\TH -ID:\Miniconda3\envs\deepwave\lib\site-packages\torch\include\THC "-IC:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.6\include" -ID:\Miniconda3\envs\deepwave\include -ID:\Miniconda3\envs\deepwave\Include "-IC:\Program Files (x86)\Microsoft Visual Studio\2019\BuildTools\VC\Tools\MSVC\14.29.30133\include" "-IC:\Program Files (x86)\Windows Kits\10\include\10.0.19041.0\ucrt" "-IC:\Program Files (x86)\Windows Kits\10\include\10.0.19041.0\shared" "-IC:\Program Files (x86)\Windows Kits\10\include\10.0.19041.0\um" "-IC:\Program Files (x86)\Windows Kits\10\include\10.0.19041.0\winrt" "-IC:\Program Files (x86)\Windows Kits\10\include\10.0.19041.0\cppwinrt" -Xcudafe --diag_suppress=dll_interface_conflict_dllexport_assumed -Xcudafe--diag_suppress=dll_interface_conflict_none_assumed -Xcudafe --diag_suppress=field_without_dll_interface -Xcudafe --diag_suppress=base_class_has_different_dll_interface -Xcompiler /EHsc -Xcompiler /wd4190 -Xcompiler /wd4018 -Xcompiler /wd4275 -Xcompiler /wd4267 -Xcompiler /wd4244 -Xcompiler /wd4251 -Xcompiler /wd4819 -Xcompiler /MD -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__-D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr --restrict -O3 --use_fast_math -DTORCH_API_INCLUDE_EXTENSION_H -DTORCH_EXTENSION_NAME=scalar1d_gpu_iso_4_float -D_GLIBCXX_USE_CXX11_ABI=0 -gencode=arch=compute_86,code=compute_86 -gencode=arch=compute_8
6,code=sm_86 --use-local-env

C:\Users\dell\AppData\Local\Temp\pip-install-hyawlsyh\deepwave_c21d48d98eb94f2487b9e27bb82533bf\deepwave\scalar\scalar.h(203): fatal error C1189: #error: "Must specify the dimension, e.g. -D DEEPWAVE_DIM=1"
scalar_gpu.cu
error: command 'C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.6\bin\nvcc.exe' failed with exit code 2
[end of output]

note: This error originates from a subprocess, and is likely not a problem with pip.
error: legacy-install-failure

× Encountered error while trying to install package.
╰─> deepwave

note: This is an issue with the package mentioned above, not pip.
hint: See above for output from the failure.

[ar4]

Hi @KELEN, I'm very glad to hear that Deepwave works well for you on the CPU. Thank you for reporting the problem you encountered when trying to install the GPU version. The error that you received is strange. It seems that the "DEEPWAVE_DIM" macro is not defined, but I don't know why. I am currently re-writing Deepwave in a way that I hope will eliminate compilation issues (the main source of problems encountered by users) and to also add some new features. I have just started this rewrite, however, so I expect that it will be a few weeks before it is ready. Would it be possible for you to wait until then, or do you need the GPU version to work urgently?

…

-Alan

[YingchangLiu]

Thanks for reply, Alan. I can wait it. The problem just occers on my Windows. Sometimes I use Google's colab and deepwave-gpu works well there. I guess there are some different rules of cuda between Colab (Linux) and Windows. I haven't test it on my linux laptop. Laptops have lower computing performance, you know.

You are doing a great work. Waiting for your new deepwave.

[YingchangLiu]

Today I can determine the problem is caused by windows. I just installed deepwave-gpu on archlinux successfully.
We have three windows computers with the same error and it doesn't appear on linux.

[ar4]

That's good news! Hopefully the next release will work just as easily on Windows.

[ar4]

Hello again,

It took longer than expected to release the new version. After much experimentation, I unfortunately decided that the TorchScript implementation that I hoped to use, which would have avoided the necessity for users to compile code and so would have avoided many problems such as those that you experienced on Windows, was not fast enough. The new version thus still uses C++/CUDA and so needs to be compiled. I changed the installation method slightly, so it might cause fewer problems, but I expect that it is probably still easiest to use it with Linux. I'm sorry that I don't have better news.

[YingchangLiu]

Hi Alan,

The new release of deepwave is a good news.

I have installed a debian for wsl2 on our workstation. So the compilation doesn't effect me anymore. Maybe I should convince my mates to use more Linux for better programming.

Thanks for your attention to this issue.